Constraining free riding in public goods games: designated solitary punishers can sustain human cooperation

 
CONTINUE READING
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

                                                                                                        Proc. R. Soc. B (2009) 276, 323–329
                                                                                                                 doi:10.1098/rspb.2008.1082
                                                                                                         Published online 23 September 2008

      Constraining free riding in public goods games:
        designated solitary punishers can sustain
                    human cooperation
                    Rick O’Gorman1,*, Joseph Henrich2 and Mark Van Vugt3
              1
             Psychology Group, Sheffield Hallam University, Collegiate Crescent Campus, Sheffield S10 2BP, UK
  2
   Psychology and Economics Departments, University of British Columbia, Vancouver, British Columbia, Canada V6T 1Z4
                         3
                          Psychology Department, University of Kent, Canterbury CT2 7NP, UK
        Much of human cooperation remains an evolutionary riddle. Unlike other animals, people frequently
        cooperate with non-relatives in large groups. Evolutionary models of large-scale cooperation require not just
        incentives for cooperation, but also a credible disincentive for free riding. Various theoretical solutions have
        been proposed and experimentally explored, including reputation monitoring and diffuse punishment.
        Here, we empirically examine an alternative theoretical proposal: responsibility for punishment can be
        borne by one specific individual. This experiment shows that allowing a single individual to punish
        increases cooperation to the same level as allowing each group member to punish and results in greater
        group profits. These results suggest a potential key function of leadership in human groups and provides
        further evidence supporting that humans will readily and knowingly behave altruistically.
                             Keywords: cooperation; free riding; punishment; altruism; leadership

1. INTRODUCTION                                                                 Observed hunter-gatherer groups adopt various mech-
In recent years, there has been a spate of papers providing                     anisms to ensure cooperation, and leadership is one such
evidence for various mechanisms to coax cooperation out                         mechanism that both integrates with humanity’s primate
of groups of individuals (Rockenbach & Milinski 2006;                           heritage and offers a mechanism for groups to coordinate
Sigmund 2007). It is to state the obvious that humans can                       activity (Brown 1991; Boehm 1999; Van Vugt 2006).
cooperate readily in extraordinary numbers (Smirnov                             Models in economics (Hirshleifer & Rasmusen 1989) and
et al. 2007) and that this cooperation often provides                           evolutionary biology (Boyd & Richerson 1992) indicate
public goods, despite the risk of free riding (Andreoni                         that evolution can favour a single punisher per social
1988; Fehr & Fischbacher 2003). Much of the recent                              group and that the actions of this one punisher can
empirical work on the puzzling aspects of human                                 efficiently galvanize group cooperation. This solution is
cooperation have focused on testing evolutionary models                         particularly interesting since it lacks the second-order free
of diffuse or altruistic punishment (Boyd & Richerson                           rider problem—which has been the central focus of much
1992; Henrich & Boyd 2001; Boyd et al. 2003), in which                          theoretical effort—and it avoids the problem of uncoordi-
many individuals share the burden of punishing non-                             nated over punishment.
cooperators (Sober & Wilson 1998; Fehr & Gächter 2002;                             Our experimental findings confirm that (i) when placed
Fehr et al. 2002; Fehr & Fischbacher 2003).                                     in the sole punisher role, individuals will punish sufficiently
   However, since recent work has shown a lack of                               to sustain cooperation, (ii) others will respond by increasing
motivation for costly punishment in some otherwise                              cooperative contributions and (iii) a single punisher can
cooperative societies (Henrich et al. 2006)—perhaps                             sustain levels of cooperation comparable with that
because the solutions have not addressed the problem of                         maintained by diffuse punishment ( Yamagishi 1986;
second-order free riding—and a possible taste for counter-                      Ostrom et al. 1994; Fehr & Gächter 2002) and at more
vailing anti-social punishment (Herrmann et al. 2008),                          profitable levels, since punishing efforts are not unnecess-
it seems plausible that different mechanisms may                                arily duplicated. Such findings suggest that in the smaller
stabilize cooperation in different ways in different                            scale societies that have dominated human evolutionary
populations. We explore a solution to n-person co-                              history (as well as in the smaller groups of contemporary
operation in which a designated individual is responsible                       societies), the single punisher solution may have been an
for punishment, contrasting with prior research in this                         important means of maintaining cooperation. In such
area (but see Carpenter 2007). Over the course of human                         groups, single punishers may even be a superior mech-
evolution, individuals in groups capable of motivating                          anism, compared with diffuse punishment systems.
cooperation would have gained an adaptive advantage.

                                                                                2. MATERIAL AND METHODS
* Author for correspondence (rogorman@alumni.binghamton.edu).                   (a) Participants
All authors contributed equally to this work.                                   One hundred and thirty six participants (35% male) who
Electronic supplementary material is available at http://dx.doi.org/10.         were undergraduate students from the University of Kent at
1098/rspb.2008.1082 or via http://journals.royalsociety.org.                    Canterbury were recruited from across the campus by way of

Received 4 August 2008
Accepted 5 September 2008                                                 323                           This journal is q 2008 The Royal Society
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

324 R. O’Gorman et al.          Solitary punishers constrain free riding

a job advertisement service. Six experimental sessions               anonymous. Investment decisions were made simultaneously,
took place with 20–24 participants per session. The                  after which information was provided on the investments of
sessions lasted approximately 1 hour and the average                 other group members. In the second-segment punishment
earning for participants was UK£5.47. Each monetary unit             conditions, individuals could simultaneously make deductions
(MU) earned during the session equated to UK£0.01.                   from each other by paying a fee, drawn from their earnings for
                                                                     that round, up to a fee maximum of 10 MUs per punished
(b) Design and procedure                                             member (the deduction was equivalent to three times the fee).
Initially, participants were informed, by way of a projected         For the one-punishment condition, one member per group
presentation, of the procedure of the experiment (including          was randomly selected after each investment phase to make
that assignment to groups was random and occurred each               deductions, whereas in the all-punishment condition, all
round, interactions were anonymous, the amount of endow-             individuals could make deductions. We conducted the public
ment, how it could be invested, how pay-offs were allocated          goods game for six rounds in each condition, so that
and how they would be paid), examples of different                   participants played a total of 12 rounds over two segments to
contribution patterns and the corresponding pay-offs, and            avoid one-shot effects and to examine participants’ behaviour
the use of the computer software. Participants were not              over a series of games. With each participant acting as his or
informed at the beginning of the first segment that there            her own control and with a fixed order, we could compare
were two segments. After the presentation of the instru-             between conditions.
ctions, participants were tested on their understanding                  During the experiment, participants received no infor-
of the pay-off procedure. All participants showed                    mation other than that of the contributions made by each of
satisfactory comprehension.                                          the other group members to the group fund and, in the
    For those in punishment conditions, further instruction          punishing conditions after punishing occurred, of the level of
was provided prior to the commencement of the second                 deductions made from their own account only. Participants
segment while those continuing with a second control                 were located in a large computer laboratory and were spaced
condition received a brief refresher. Instructions relating to       apart such that no one could see another participant’s screen.
the making of deductions did not make any suggestion as to           After completing the public goods games, participants
how such deductions could be used, or whether they should            completed online and paper questionnaires to assess their
be used. Participants were simply informed that such                 attitudinal and emotional responses to the experiment and
deductions would be possible for the second segment and it           their interactions in the games, group identity and a number
was explained how to make such deductions, should                    of other measures not reported here.
participants wish to use such a facility. If a participant
queried the purpose, then he or she was simply told that it was
an option that would be available and it was up to him or her        3. RESULTS
how it could be used.                                                The average contribution made by participants across all
    We used a modified methodology ( Fehr & Gächter 2002)           sessions and rounds was 8.28 MUs (s.d.Z6.55). For
of a public goods experiment that had real monetary earnings         analysis, we used generalized estimating equations
at stake run on networked PCs using Z-TREE software                  (GEEs), available in SPSS v. 15, which uses robust
( Fischbacher 2007). All participants completed a two-               (Huber–White) errors to correct for lack of independence
segment experiment with an initial no-punishing control              in the data. Because participants interacted with each
segment followed by a second segment of either a further             other within sessions, this represents a conservative
control condition (no-punishment), a condition with punish-          approach to analysis (we also performed a non-parametric
ment permitted for all group members (all-punishment), or a          analysis, which yielded qualitatively similar results;
condition with only one individual permitted to punish               however, GEE allows for more powerful analysis and is
(one-punishment); therefore, all participants acted as their         what we report here). We present our analysis firstly of the
own controls and partook in only one of three conditions. We         contribution data, then of the profit data and finally of the
did not counterbalance as, firstly, Fehr & Gächter (2002)           punishing data.
showed that there was no order effect for not punishing
versus punishing and, secondly, our focus was on comparison          (a) Analysis of contributions
between the two punishment conditions.                               GEEs for the contribution data used a first-order
    In all conditions, participants played the same public           autoregressive working correlation matrix, due to corre-
goods game: assigned to groups of four, participants were            lations between adjacent rounds’ contributions and a
allocated an endowment of 20 MUs, of which they could                normal/identity link. The data from segment 2 of the study
invest any amount into a group fund and retain the                   concerns our primary hypothesis that the one-punishment
remainder. Each MU invested in the group fund yielded a              condition would increase contributions above the
pay-off of 0.5 MU to each group member, irrespective of who          control condition (see figure 1 for mean contributions
invested. That is, each MU invested in the group fund was            and 95% confidence intervals for the three conditions
doubled and then divided into four equal shares. Thus,               over the six rounds). We examined the effects of condition,
participants would always be better off contributing nothing         round and sex on contributions. There is a main effect on
to the group fund as the return was less than the investment.        segment 2 contributions (Wald c22Z10.41, pZ0.005).
However, if every member invested their full endowment,              Regression values (derived from the GEE model; see the
then each member would earn 40 MUs, a profit of 20.                  electronic supplementary material S1) show that both all-
    In each round, groups were randomly formed so that               punishment and one-punishment differ significantly from
participants never knew with whom they were interacting              the control group (all-punishment versus control: BZ6.20,
(‘stranger protocol’ in the economics literature), thus control-     s.e.Z1.32, p!0.001; one-punishment versus control:
ling for reputation and reciprocity effects. All interactions were   BZ5.47, s.e.Z1.19, p!0.001) while all-punishment and

Proc. R. Soc. B (2009)
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

                                                                             Solitary punishers constrain free riding        R. O’Gorman et al.   325

                               14                                                                            40

                               12                                                                            35
    group contribution means

                                                                                                             30
                               10

                                                                                        group profit means
                                                                                                             25
                               8
                                                                                                             20
                               6
                                                                                                             15
                               4
                                                                                                             10
                               2
                                                                                                              5

                               0                                                                              0
                                    1   2     3      4      5      6                                              1   2   3      4   5     6
                                               round                                                                       round
Figure 1. Mean contributions of MUs to the group fund by                            Figure 2. Mean profits (MUs) for participants in segment 2
participants in segment 2 with 95% confidence intervals                             with 95% confidence intervals indicated by error bars.
indicated by error bars. Circles, all-punishment; squares,                          Circles, all-punishment; squares, one-punishment; triangles,
one-punishment; triangles, no-punishment.                                           no-punishment.

one-punishment do not appear to significantly differ                                (b) Analysis of profits
                                                                                    Differences in segment 1 profits due to condition and
(BZK0.73, s.e.Z1.32, pZ0.579; obtained by switching
                                                                                    round necessarily follow contribution differences in the
the reference category from control to all-punishment).
                                                                                    same study segment and so, not surprisingly, are signifi-
    Additionally, a main effect for round approaches
                                                                                    cant (condition: Wald c22Z25.68, p!0.001; round: Wald
significance (Wald c25Z10.74, pZ0.057) and there is an
                                                                                    c25Z33.68, p!0.001), though there is no effect for sex or
interaction between manipulation and rounds (Wald
                                                                                    interactions. As above, for analysis of the profit data from
c210Z25.29, pZ0.005), reflecting the decrease in contri-
                                                                                    segment 2 (see figure 2 for mean profits and 95%
butions in the no-punishment condition in contrast to the                           confidence intervals), we use segment 1 contributions as
more stable contributions in the other two conditions                               a covariate. There is a main effect for condition (Wald
(figure 1). Contributions in the control condition decreased                        c22Z144.79, p!0.001) and an interaction between con-
significantly across the six rounds in segment 2 (rounds                            dition and round (Wald c210Z42.10, p!0.001), though no
regressed on contributions with robust errors, BZK1.08,                             main effects for round or sex, nor are there interaction
s.e.Z0.22, p!0.001), whereas contributions in the two                               effects. Regression values (as earlier, derived from the
punishing conditions remained relatively constant                                   GEE model, see the electronic supplementary material S2)
(all-punishment BZK0.12, s.e.Z0.21, pZ0.570; one-                                   show that all three conditions differ (all-punishment
punishment BZ0.03; s.e.Z0.19, pZ0.868). There is no                                 versus control: BZK11.55, s.e.Z2.03, p!0.001; one-
effect for sex, nor is there an interaction ( pO0.370). Our                         punishment versus control: BZK5.48, s.e.Z1.79,
findings support the hypothesis that, under these con-                              pZ0.002; all-punishment versus one-punishment:
ditions, a single individual operating as the sole punisher in                      BZ6.07, s.e.Z2.32, pZ0.009; the latter is again obtained
a group can improve contributions relative to a control                             by switching the reference category from control to
condition without punishment and matches the effect                                 all-punishment).
produced by allowing everyone to punish.                                                The lower mean values for the punishment conditions
    We should note that there are differences between                               are primarily due to the cost of punishing and deductions,
conditions in contributions in segment 1 (Wald c22Z19.59,                           relative to the control condition. However, it is worth
p!0.001), possibly due to participants attending with an                            noting that, whereas the slopes of the punishment
understanding of the experiment, but these initial                                  conditions appear stable relative to rounds (rounds reg-
differences disappear after the six rounds (round 1: Wald                           ressed on contributions with robust errors, all-punishment
c22Z16.72, p!0.001; round 6: Wald c22Z1.80, pZ0.408).                               BZ0.01, s.e.Z0.46, pZ0.980; one-punishment BZ0.21;
However, this change is not reflected in a significant                              s.e.Z0.34, pZ0.542), the control condition’s slope is not
interaction (Wald c210Z4.68, pZ0.912), though there is a                            (BZK1.08, s.e.Z0.23, p!0.001), suggesting that both
main effect for round (Wald c25Z33.09, p!0.001). Finally,                           punishment conditions would be likely to be more
there is no difference due to sex of participant (Wald c21Z                         profitable than the control condition in the long run.
2.91, pZ0.088), nor did sex interact with either condition                          Importantly, the one-punishment condition has an
or round ( pO0.711). The lack of a significant difference                           advantage over the all-punishment condition due to
between conditions in round 6 of segment 1 suggests that                            lower total costs incurred by group members and this is
                                                                                    reflected in its higher profit levels (figure 2).
any initial differences in contribution levels between
conditions had been eliminated by the end of segment 1,
but to control for differences in baseline contribution                             (c) Analysis of punishment
dispositions, we used participants’ average contributions in                        Looking at punishing, overall participants in the all-
segment 1 as a covariate in the analysis of segment 2 data.                         punishment condition punished on 38.9 per cent of

Proc. R. Soc. B (2009)
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

326 R. O’Gorman et al.                         Solitary punishers constrain free riding

                  (a)                    1.0                                        (b)                    1.0

                                         0.8                                                               0.8

                  proportion punishing

                                                                                     proportion punished
                                         0.6                                                               0.6

                                         0.4                                                               0.4

                                         0.2                                                               0.2

                                          0                                                                 0

                  (c)                    16                                         (d)                    25
                                         14
                                                                                                           20
                                         12

                                                                                     mean punishment
                                                                                      incurred (MUs)
                  mean punishment
                  expended (MUs)

                                         10                                                                15
                                          8
                                          6                                                                10

                                          4
                                                                                                            5
                                          2
                                          0                                                                 0
                                                 1    2    3   4      5    6                                     1   2   3   4   5   6
                                                           round                                                         round
Figure 3. (a) There were more punishers in the one-punishment condition who punished at least once per round than in the all-
punishment condition, (b) although punishers in the one-punishment condition did not punish as many group members.
(c) One-punishment punishers did, however, expend greater resources to punish, (d ) resulting in a similar level of penalties
being incurred within each punishment condition when considered over the six rounds in segment 2 (circles, all-punishment;
triangles, one-punishment).

opportunities to do so whereas punishers in the one-                                pZ0.001). However, when the receiver’s contribution was
punishment condition did so on 56.6 per cent of                                     greater, higher punisher contributions are associated with
opportunities. Per round, the proportion of participants                            lower punishments. In the one-punishment condition, the
who punished in the one-punishment condition (i.e.                                  pattern was similar irrespective of whether the punisher’s or
punished at least once) was greater than the proportion                             target’s contribution was greater. In both cases, we found
in the all-punishment condition (figure 3a). Fewer                                  that higher punishment is associated with lower contri-
participants were punished in the one-punishment con-                               butions by the target.
dition (figure 3b) but that condition’s punishers made                                  Thus, it appears that while participants in diffuse puni-
greater deductions (figure 3c), although the total incurred                         shment situations attend to both their own contribution and
punishments were not consistently harsher in either                                 that of the target, perhaps using their own contributions to
condition (figure 3d ).                                                             guide their decision on whether to punish, those in the
   When we examined possible factors that affected                                  solitary punisher condition attend only to the con-
punishment behaviour, we found that punishers in the all-                           tributions of the target, possibly focused solely on whether
punishment condition and the one-punishment condition                               contributions are maximally beneficial for the group, in
appear to be influenced by the same factors. Using a GEE                            which case any deviation from a full contribution represents
approach (using a gamma/log link, due to the positively                             an undesirable shortfall. Figure 4 shows that for both
skewed data), we examined separately for the two punish-                            punishment conditions, lower target contributions are
ment conditions the relationship of punishment levels with                          associated with higher punishment, although this pattern
the contribution of potential punishers and targets, and                            is clearer for those in the all-punishment condition
rounds. In addition, we also split our analysis between cases                       (figure 4a). However, it is worth drawing attention to the
where the potential punisher’s contribution was less than                           finding for the all-punishment data that lower punisher
the target’s contribution, versus when it was equal to, or                          contributions are associated with higher punishment when
greater than, the target’s. Examining the all-punishment                            the target has made a higher contribution. This resonates
condition (see table 1 for GEE regression parameter                                 with prior findings which suggest that some participants
estimates), when the punisher’s contribution was greater,                           may be acting spitefully (Herrmann et al. 2008). However,
we found that higher punishment is associated with                                  such interactions place together such cases where both
lower contributions by the target (BZK0.086, s.e.Z0.01,                             have contributed very little (one point versus two points),
p!0.001), while higher punisher contributions are also                              moderate amounts (10 versus 11), or substantial amounts
associated with higher punishment (BZ0.039, s.e.Z0.01,                              (19 versus 20). We would not expect the behaviour in these

Proc. R. Soc. B (2009)
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

                                                                                         Solitary punishers constrain free riding                   R. O’Gorman et al.        327

Table 1. GEE parameter estimates for regression of imposed punishment on punisher’s and target’s contributions and round in
study segment 2. (Round 6 was the reference category for the round factor. The results were analysed separately for all-
punishment and one-punishment conditions, and for when the punisher’s contribution was equal to or greater than the target’s
contribution and when the target’s contribution was greater.)

parameter                                         B               s.e.              Wald c2                         sig.       B             s.e.           Wald c2      sig.

                                                  all-punishment/punisher’s contribution greater                               all-punishment/target’s contribution greater
intercept                                            0.974     0.1865         27.264        0.000                                 0.898     0.2381          14.211        0.000
punisher’s contribution                              0.039     0.0119         10.989        0.001                              K0.071       0.0265           7.107        0.008
target’s contribution                             K0.086       0.0129         44.568        0.000                              K0.001       0.0120           0.003        0.954
round 1                                           K0.054       0.1691          0.100        0.751                                 0.241     0.2055           1.379        0.240
round 2                                           K0.340       0.1102          9.521        0.002                                 0.028     0.2515           0.012        0.911
round 3                                           K0.215       0.1678          1.638        0.201                                 0.071     0.1553           0.211        0.646
round 4                                           K0.238       0.1204          3.910        0.048                                 0.021     0.1873           0.013        0.910
round 5                                           K0.087       0.1694          0.264        0.607                                 0.269     0.2068           1.692        0.193
                                                  one-punishment/punisher’s contribution greater                               one-punishment/target’s contribution greater
intercept                                          0.683          0.1998            11.692                          0.001       0.753        0.3831         3.861        0.049
punisher’s contribution                           K0.003          0.0124             0.063                          0.802       0.015        0.0157         0.929        0.335
target’s contribution                             K0.021          0.0103             4.232                          0.040      K0.025        0.0128         3.827        0.050
round 1                                            0.038          0.2335             0.026                          0.872      K0.366        0.3119         1.378        0.240
round 2                                            0.088          0.2577             0.116                          0.733       0.112        0.3676         0.093        0.761
round 3                                            0.190          0.2586             0.539                          0.463       0.087        0.3446         0.064        0.801
round 4                                            0.175          0.2521             0.483                          0.487      K0.095        0.4381         0.047        0.829
round 5                                           K0.386          0.1725             5.017                          0.025       0.071        0.3798         0.035        0.852

                     (a) 10                                                                (b) 10
                         imposed punishment

                                                                                           imposed punishment

                                              8                                                                 8

                                              6                                                                 6

                                              4                                                                 4

                                              2                                                                 2

                                                  0                                                       0
                                                         5                                                      5
                                                  target     10                          15  20            target    10                        15 20 n
                                                         contri      15
                                                                bution    20 0   5   10
                                                                                               bu t i o n         contri 15 20 0 5 10              ibutio
                                                                                h e r c ontr i                          bution
                                                                                                                                     s h e r contr
                                                                          punis                                                pun i

Figure 4. Punishers tended to apply greater deductions for values that deviated more from higher levels of possible
contributions, though this effect is stronger in the (a) all-punishment than (b) one-punishment condition.

three examples to be similar, yet in all cases the target has                                                   study (in either punishment condition) is facilitated by the
contributed one point more. Thus, we need further study to                                                      1 : 3 ratio of the cost of punishing for the punisher to the
understand the strategies being used by participants.                                                           cost for the target. While this is a common ratio in this
Clearly, the strategies are not as straightforward as might                                                     methodology, studies have shown that lower ratios tend
be anticipated.                                                                                                 not to produce punishing behaviour sufficient to sustain
                                                                                                                cooperation ( Yamagishi 1986; Burnham & Johnson 2005;
                                                                                                                Nikiforakis & Normann 2008). However, we do not view
4. DISCUSSION                                                                                                   this as an unnecessary stumbling block. Asymmetrical
Individual contributions were significantly higher when                                                         impacts of punishment can be readily achieved in the
punishment was available as an option, with participants                                                        real world, for example, through the use of a weapon or
responding as effectively to a single individual as to all                                                      social support.
group members making deductions. Our results suggest                                                                The pattern of punishment for contribution levels
that a single punisher successfully enhances and stabilizes                                                     suggests that lower contributions tend to incur greater
group contributions, while doing so more profitably than                                                        levels of punishment. As Carpenter & Matthews (2008)
in the all-punishment condition. As punishment costs are                                                        argue, it appears that punishers are more focused on actual
lost to the system, punishments by a single punisher are                                                        contribution levels rather than deviations from the group
more coordinated and thus reduce inefficient losses. It is                                                      (or session) mean, per se. However, actual strategies in
important to note that the success of punishing in this                                                         anonymous games inevitably are likely to be complex,

Proc. R. Soc. B (2009)
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

328 R. O’Gorman et al.        Solitary punishers constrain free riding

reflecting the fact that individuals vary in their cooperative     Richerson & Boyd 2004). Groups often favour altruists
intent ( Van Lange 1999) and thus how they respond to              for the leader role (Milinski et al. 2002; Hardy & Van Vugt
both being able to ‘punish’ and being ‘punished’. Further          2006). Competition between rival groups results in selective
in-depth examination of participants’ strategies, motives          pressure for its adoption culturally or its evolution,
and goals is needed.                                               genetically. If all participants can punish each other, such
   Somewhat unexpectedly, more participants in the one-            situations risk deteriorating into retaliatory actions that do
punishment condition punished more often and more                  not just reduce benefits from joint activities but damage the
harshly than those in the all-punishment condition,                integrity of the group (Denant-Boemont et al. 2007;
incurring greater personal costs. There are a number of            Nikiforakis 2008). A designated punisher avoids these risks.
possible explanations for this result. The first is that the           The issue of anonymity and the consequential inability
study protocol with a single designated punisher may have          of punished individuals to retaliate represent a constraint
simply enhanced an experimental demand characteristic              on our argument that we provide evidence for leadership
for participants to punish, with the one-punishment                to function as a constraint on free riding. If retaliation were
condition having increased compliance due to the                   possible, a single punishing individual would be less costly
reduction of diffusion of responsibility (Latané & Darley         to retaliate against than a set of punishers. However, in this
1970), although participants would have had to incur a real        study, we seek only to demonstrate that leadership could
cost to comply (in contrast to many studies), and some             fulfil such a function successfully. In reality, a leader is not
researchers dispute whether participants do indeed                 just one individual but represents the pinnacle of a social
respond to such demands (Berkowitz & Troccoli 1986).               structure. Thus, although responsibility may lie with one
The second is that participants may indeed have not felt           individual to act, such actions nonetheless, by virtue of the
their actions were anonymous and, cued by the presence of          role, carry the support of the group, or at least a majority.
other participants, acted in a manner that they found              Additionally, the actual form of punishment varies
appropriate to enhance, or at least maintain, their                substantially, and indeed a leader may not need to be
reputation as positive group members (Haley & Fessler              the individual to actually impose the punishment, as
2005; Hagen & Hammerstein 2006). In this case,                     caricatured by Mafia films and as is very familiar to anyone
participants would have to view the incurred real costs as         working in an organization with punishment capabilities.
necessary for the (imagined) gain to reputation, and to                In the present study, the random selection of punishers in
view punishing and acting altruistically as a reputation-          the one-punishment condition served as a means to impose
enhancing behaviour rather than signalling aggression or           the role on individuals to control for other confounds.
vulnerability to exploitation. Finally, humans may altruis-        Nonetheless, future studies would do well to attend to more
tically punish for the benefit of their group ( Fehr &             realistic exploration of the role of leaders as punishers. One
Gächter 2002), at least under certain conditions. Although        interesting follow-up would be to examine a series of
participants were ostensibly anonymous, even if they were          experimental rounds, allowing participants either to
not convinced of this state, designated punishers in the           experience different regimes (no punishing, diffuse punish-
one-punishment condition knew that they alone would                ing, single punisher) or gain information on the per-
carry the costs of acting to the benefit of the group              formance of different regimes, and choose which system
(assuming that they saw punishing as such).                        to play under. This could further demonstrate the will-
   These findings may have an important implication for            ingness (or not) of individuals to operate under a designated
the study of cooperation and the functions of leadership           punisher (leader) system. Related research documenting
in humans. As noted earlier, large-scale cooperation in            cross-cultural variation in costly punishing (Henrich et al.
human groups (beyond the hunter-gatherer level) rep-               2006) suggests our findings may be constrained and it
resents an evolutionary puzzle. Diffuse punishment does
                                                                   would be worthwhile to consider whether punishing
not fully solve this issue owing to the iterated problem of
                                                                   through leadership is a cultural universal. The potential
second-order free riders. A system with a single designated
                                                                   impact of retaliation also warrants consideration.
punisher can potentially avoid this problem because there
                                                                       In smaller scale human societies, prestigious leaders
is clearer accountability. In human groups, leaders often
                                                                   can galvanize the trace of larger scale cooperation
fulfil the role of designated punishers (Heizer 1978;
                                                                   ( Johnson 2003). At least in some circumstances, individ-
Krackle 1978; Diamond 1997). Moreover, some form of
                                                                   uals respond as effectively to a single punishing individual
leadership, even if only ephemeral (Steward 1938;
                                                                   as they do to a more general punitive environment without
Johnson & Earle 2000), is a human universal and readily
                                                                   obvious negative reactions. Consistent with the existing
emerges in ad hoc laboratory groups ( Van Vugt 2006).
                                                                   theoretical work (Boyd & Richerson 1992), our research
   Of course, such a leadership role is potentially costly to
                                                                   suggests that human psychology may have evolved to
the individual who occupies it. There is both the energy
                                                                   recognize situations in which a single motivated leader can
budget of punishing and the incumbent costs of self-
                                                                   enforce cooperation ( Van Vugt et al. 2008).
defence by the target or retaliation. Why would individuals
take on this role? There may be compensatory benefits              This research complied with ethical standards for the
for acting as a leader. Some individuals more readily fulfil       treatment of participants and received institutional ethical
this role than others based on heritable differences in            approval.
personality (Hogan et al. 1994). In human societies, leaders
                                                                   We acknowledge the support by the British Academy for this
acquire status and prestige ( Van Vugt 2006), which may
                                                                   work. We would like to thank anonymous reviewers for their
translate into increased reproductive success (Henrich &           helpful comments.
Gil-White 2001; Fieder et al. 2005). Alternatively, group-
level selection could facilitate leadership emergence, either      Author contributions. The authors declare that they have no
by genetic or cultural mechanisms (Sober & Wilson 1998;            competing financial interests.

Proc. R. Soc. B (2009)
Downloaded from http://rspb.royalsocietypublishing.org/ on May 10, 2015

                                                             Solitary punishers constrain free riding   R. O’Gorman et al.     329

REFERENCES                                                          Henrich, J. & Gil-White, F. J. 2001 The evolution of prestige:
Andreoni, J. 1988 Why free ride? Strategies and learning in            freely conferred deference as a mechanism for enhancing
   public good experiments. J. Public Econ. 37, 291–304.               the benefits of cultural transmission. Evol. Hum. Behav.
   (doi:10.1016/0047-2727(88)90043-6)                                  22, 165–196. (doi:10.1016/S1090-5138(00)00071-4)
Berkowitz, L. & Troccoli, B. T. 1986 An examination of the          Henrich, J. et al. 2006 Costly punishment across human
   assumptions in the demand characteristics thesis: with              societies. Science 312, 1767–1770. (doi:10.1126/science.
   special reference to the Velten mood induction procedure.           1127333)
   Motiv. Emot. 10, 337–349. (doi:10.1007/BF00992108)               Herrmann, B., Thöni, C. & Gächter, S. 2008 Antisocial
Boehm, C. 1999 Hierarchy in the forest. Cambridge, MA:                 punishment across societies. Science 319, 1362–1367.
   Harvard University Press.                                           (doi:10.1126/science.1153808)
Boyd, R. & Richerson, P. J. 1992 Punishment allows the              Hirshleifer, D. & Rasmusen, E. 1989 Cooperation in a repeated
   evolution of cooperation (or anything else) in sizable              Prisoners’ Dilemma with ostracism. J. Econ. Behav. Organ.
   groups. Ethol. Sociobiol. 13, 171–195. (doi:10.1016/0162-           12, 87–106. (doi:10.1016/0167-2681(89)90078-4)
   3095(92)90032-Y)                                                 Hogan, R., Curphy, G. J. & Hogan, J. 1994 What we know
Boyd, R., Gintis, H., Bowles, S. & Richerson, P. J. 2003 The           about leadership. Am. Psychol. 49, 493–504. (doi:10.1037/
   evolution of altruistic punishment. Proc. Natl Acad. Sci.           0003-066X.49.6.493)
   USA 100, 3531–3535. (doi:10.1073/pnas.0630443100)                Johnson, A. 2003 Families of the forest: Matsigenka Indians
Brown, D. 1991 Human universals. Boston, MA: McGraw-                   of the Peruvian Amazon. Berkeley, CA: University of
   Hill.                                                               California.
Burnham, T. C. & Johnson, D. D. P. 2005 The biological and          Johnson, A. & Earle, T. 2000 The evolution of human societies.
   evolutionary logic of human cooperation. Anal. Kritik 27,           Stanford, CA: Stanford University Press.
   113–135.                                                         Krackle, W. H. 1978 Force and persuasion: leadership in an
Carpenter, J. P. 2007 Punishing free-riders: how group size            Amazonian Society. Chicago, IL: University of Chicago
   affects mutual monitoring and the provision of public               Press.
   goods. Games Econ. Behav. 60, 31–51. (doi:10.1016/j.geb.         Latané, B. & Darley, J. M. 1970 The unresponsive bystander:
   2006.08.011)                                                        why doesn’t he help? New York, NY: Appleton-Crofts.
Carpenter, J. P. & Matthews, P. H. 2008 What norms trigger          Milinski, M., Semmann, D. & Krambeck, H. 2002 Donors to
   punishment. Working paper obtained from http://commu-               charity gain in both indirect reciprocity and political
                                                                       reputation. Proc. R. Soc. B 269, 881–883. (doi:10.1098/
   nity.middlebury.edu/wjcarpent/papers.html on 20 June
                                                                       rspb.2002.1964)
   2008.
                                                                    Nikiforakis, N. 2008 Punishment and counter-punishment in
Denant-Boemont, L., Masclet, D. & Noussair, C. N. 2007
                                                                       public good games: can we really govern ourselves? J. Public
   Punishment, counterpunishment and sanction enforce-
                                                                       Econ. 92, 91–112. (doi:10.1016/j.jpubeco.2007.04.008)
   ment in a social dilemma experiment. Econ. Theor. 33,
                                                                    Nikiforakis, N. & Normann, H.-T. 2008 A comparative statics
   145–167. (doi:10.1007/s00199-007-0212-0)
                                                                       analysis of punishment in public goods experiments. Exp.
Diamond, J. 1997 Guns, germs, and steel: the fates of human
                                                                       Econ 11, 358–369. (doi:10.1007/s10683-007-9171-3)
   societies. New York, NY: Norton.
                                                                    Ostrom, E. R., Gardner, R. & Walker, J. M. 1994 Rules,
Fehr, E. & Fischbacher, U. 2003 The nature of human
                                                                       games, and common-pool resources. Ann Arbor, MI:
   altruism. Nature 425, 785–791. (doi:10.1038/nature02043)
                                                                       University of Michigan Press.
Fehr, E. & Gächter, S. 2002 Altruistic punishment in
                                                                    Richerson, P. J. & Boyd, R. 2004 Not by genes alone: how
   humans. Nature 415, 137–140. (doi:10.1038/415137a)                  culture transformed human evolution. Chicago, IL:
Fehr, E., Fischbacher, U. & Gächter, S. 2002 Strong                   University of Chicago Press.
   reciprocity, human cooperation, and the enforcement of           Rockenbach, B. & Milinski, M. 2006 The efficient interaction
   social norms. Hum. Nat. 13, 1–25. (doi:10.1007/s12110-              of indirect reciprocity and costly punishment. Nature 444,
   002-1012-7)                                                         718–723. (doi:10.1038/nature05229)
Fieder, M., Huber, S., Bookstein, F., Iber, K., Schäfer, K.,       Sigmund, K. 2007 Punish or perish? Retaliation and
   Wallner, B. & Winckler, G. 2005 Status and reproduction             collaboration among humans. Trends Ecol. Evol. 22,
   in humans: new evidence for the validity of evolutionary            593–600. (doi:10.1016/j.tree.2007.06.012)
   explanations on basis of a university sample. Ethology 111,      Smirnov, O., Arrow, H., Kennett, D. & Orbell, J. 2007 Ancestral
   940–950. (doi:10.1111/j.1439-0310.2005.01129.x)                     war and the evolutionary origins of “heroism”. J. Polit.
Fischbacher, U. 2007 Z-TREE: Zurich toolbox for ready-made             69, 927–940. (doi:10.1111/j.1468-2508.2007.00599.x)
   economic experiments. Exp. Econ. 10, 171–178. (doi:10.           Sober, E. & Wilson, D. S. 1998 Unto others: the evolution and
   1007/s10683-006-9159-4)                                             psychology of unselfish behavior. Cambridge, MA: Harvard
Hagen, E. H. & Hammerstein, P. 2006 Game theory and                    University Press.
   human evolution: a critique of some recent interpretations       Steward, J. 1938 Basin–plateau aboriginal sociopolitical groups.
   of experimental games. Theor. Popul. Biol. 69, 339–348.             Washington, DC: Bureau of American Ethnology.
   (doi:10.1016/j.tpb.2005.09.005)                                  Van Lange, P. A. M. 1999 The pursuit of joint outcomes and
Haley, K. J. & Fessler, D. M. T. 2005 Nobody’s watching?               equality in outcomes: an integrative model of social value
   Subtle cues affect generosity in an anonymous economic              orientation. J. Pers. Soc. Psychol. 77, 337–349. (doi:10.
   game. Evol. Hum. Behav. 26, 245–256. (doi:10.1016/                  1037/0022-3514.77.2.337)
   j.evolhumbehav.2005.01.002)                                      Van Vugt, M. 2006 Evolutionary origins of leadership and
Hardy, C. & Van Vugt, M. 2006 Nice guys finish first: the              followership. Pers. Soc. Psychol. Rev. 10, 354–371. (doi:10.
   competitive altruism hypothesis. Pers. Soc. Psychol. Bull.          1207/s15327957pspr1004_5)
   32, 1402–1413. (doi:10.1177/0146167206291006)                    Van Vugt, M., Hogan, R. & Kaiser, R. 2008 Leadership,
Heizer, R. (ed.) 1978 Handbook of North American Indians:              followership, and evolution: some lessons from the
   California, 8. Washington, DC: Smithsonian Institution.             past. Am. Psychol. 63, 182–196. (doi:10.1037/0003-
Henrich, J. & Boyd, R. 2001 Why people punish defectors:               066X.63.3.182)
   weak conformist transmission can stabilize costly enforce-       Yamagishi, T. 1986 The provision of a sanctioning system as a
   ment of norms in cooperative dilemmas. J. Theor. Biol.              public good. J. Pers. Soc. Psychol. 51, 110–116. (doi:10.
   208, 79–89. (doi:10.1006/jtbi.2000.2202)                            1037/0022-3514.51.1.110)

Proc. R. Soc. B (2009)
You can also read